Multivariable control method for urban sewage treatment process, storage medium and server

The construction of an agent through deep reinforcement learning algorithm solves the problems of insufficient accuracy and poor adaptability in sewage treatment, and realizes accurate coordinated control of dissolved oxygen and nitrate nitrogen concentrations, which improves the control accuracy of sewage treatment process and stability.

CN119065251APending Publication Date: 2024-12-03BEIJING UNIV OF TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411186761.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional PID control has problems such as insufficient control accuracy and poor adaptability during sewage treatment, and it is difficult to effectively coordinate the control of dissolved oxygen concentration and nitrate nitrogen concentration, which affects the effluent water quality and system operation energy consumption.

Method used

The deep reinforcement learning algorithm is used to build DO agents and NO agents. Based on simulation models and actual data training, the controller parameters of dissolved oxygen and nitrate nitrogen concentration are adjusted to achieve accurate coordinated control.

Benefits of technology

It improves the control accuracy of dissolved oxygen and nitrate nitrogen concentrations, can stably track the set value, quickly adapt to the changes in expected value, and improves the control accuracy and adaptability of the sewage treatment process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065251A_ABST
    Figure CN119065251A_ABST
Patent Text Reader

Abstract

The invention discloses a multivariable control method for an urban sewage treatment process, and the method comprises the steps: constructing a sewage treatment process control environment, determining a state space function, an action space function and a reward function, constructing a DO intelligent agent and an NO intelligent agent based on a deep reinforcement learning algorithm, and carrying out the training, according to the trained DO intelligent agent and NO intelligent agent, a second controller is obtained and used for controlling the nitrate nitrogen concentration and the dissolved oxygen concentration, and stable tracking of the set values of the dissolved oxygen concentration and the nitrate nitrogen concentration is achieved; even if the expected values of the dissolved oxygen concentration and the nitrate nitrogen concentration are changed, the dissolved oxygen concentration and the nitrate nitrogen concentration can be quickly stabilized near new expected values, and the accuracy of controlling the dissolved oxygen concentration and the nitrate nitrogen concentration is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the technical field of sewage treatment, and in particular, relates to a multivariable control method for a municipal sewage treatment process. Background Art

[0002] Urban sewage contains many harmful elements and pollutants. If it is directly discharged into the water without proper treatment, it will cause serious pollution to water resources and destroy the balance of water quality and aquatic ecosystems. Effective sewage treatment measures can remove these pollutants, reduce water pollution, and ensure the sustainable stability of groundwater resources and aquatic ecosystems. Excessive organic matter and nutrients in urban sewage are the main causes of eutrophication of water bodies, which will cause excessive algae reproduction and death of aquatic organisms, thereby destroying the balance of water ecology. Through scientific sewage treatment, these organic matter and nutrients can be removed, the risk of eutrophication of water bodies can be reduced, and the health of water ecosystems can be maintained. Therefore, obtaining clean and safe water resources and timely and effective sewage treatment have become important issues that need to be solved in the current society. The development prospects of sewage treatment technology are broad. Researching and developing efficient sewage treatment control technology to achieve precise control of sewage treatment process and energy saving and consumption reduction is an inevitable trend for the future development of the sewage treatment industry.

[0003] In the sewage treatment process, dissolved oxygen concentration and nitrate nitrogen concentration play a vital role in sewage treatment. Too high dissolved oxygen concentration will cause excess dissolved oxygen to enter the anoxic zone through internal recirculation, thereby inhibiting the denitrification reaction. Too low dissolved oxygen concentration will make the nitrification reaction insufficient, and furthermore, affect the denitrification reaction. In addition, too low nitrate nitrogen concentration will also affect the denitrification reaction. Therefore, the control goal of sewage treatment is to ensure that the effluent water quality meets the standard by controlling the dissolved oxygen concentration and nitrate nitrogen concentration. At the same time, the dissolved oxygen concentration and nitrate nitrogen concentration also have a direct impact on the effluent water quality and system operation energy consumption of the sewage treatment plant. Therefore, the precise coordinated control of dissolved oxygen concentration and nitrate nitrogen concentration is the focus of the sewage treatment process.

[0004] Traditional PID control technology is widely used in sewage treatment plants due to its simple structure, high reliability and easy implementation. However, in sewage treatment systems, due to the nonlinearity, time-varying characteristics and significant hysteresis of the control object, as well as the significant influence of ambient temperature and other external factors, it becomes difficult to establish an accurate mathematical model, and thus it is difficult to determine the optimal controller parameters. This leads to problems such as insufficient control accuracy and poor adaptability when using traditional PID control. Therefore, when applying traditional PID control methods to sewage treatment processes, their limitations become apparent. Summary of the invention

[0005] The purpose of this application is to provide a multivariable control method for an urban sewage treatment process, aiming to solve the problem of insufficient control accuracy in sewage treatment.

[0006] A first aspect of an embodiment of the present application provides a multivariable control method for a municipal sewage treatment process, comprising:

[0007] Constructing a sewage treatment process control environment according to a simulation model of an activated sludge method sewage treatment process and a first controller;

[0008] determining a state space function according to control errors of dissolved oxygen concentration and nitrate nitrogen concentration, determining an action space function according to action space parameters of dissolved oxygen of the first controller and action space parameters of nitrate nitrogen of the first controller, and determining a reward function according to a difference between an actual measured value and an expected value of dissolved oxygen concentration and a difference between an actual measured value and an expected value of nitrate nitrogen concentration;

[0009] Constructing a DO agent and a NO agent based on a deep reinforcement learning algorithm, wherein the DO agent is used to adjust the parameters of the first controller for controlling the dissolved oxygen concentration, and the NO agent is used to adjust the parameters of the first controller for controlling the nitrate nitrogen concentration;

[0010] Based on the sewage process control environment, training the DO agent and the NO agent; and

[0011] A second controller is obtained according to the trained DO agent and NO agent, and the second controller is applied to the simulation model of the activated sludge wastewater treatment process to control the nitrate nitrogen concentration and the dissolved oxygen concentration.

[0012] In one embodiment, the construction of a sewage treatment process control environment based on a simulation model of an activated sludge sewage treatment process and a first controller includes:

[0013] According to the simulation benchmark model No. 1 BSM1 in the activated sludge wastewater treatment process, a biochemical reaction tank and a secondary sedimentation tank are constructed; wherein the biochemical reaction tank includes a first unit, a second unit, a third unit, a fourth unit and a fifth unit;

[0014] A first controller is constructed as a dissolved oxygen concentration controller of the fifth unit of the biochemical reaction tank and a nitrate nitrogen concentration controller of the second unit of the biochemical reaction tank; wherein the first controller is a MADRL controller.

[0015] In one embodiment, the determining of the state space function according to the control error of the dissolved oxygen concentration and the nitrate nitrogen concentration comprises:

[0016] The state space function s(k) is the state space o of the DO agent. DO(k) and the state space o of the NO agent NO (k) is composed of, satisfying s(k) = [o DO (k) T ,o NO (k) T ] T ;

[0017] The state space o of the DO agent DO (k) Control error by dissolved oxygen concentration constitute, satisfy

[0018] The state space of the NO agent is controlled by the nitrate nitrogen concentration error constitute, satisfy at the same time

[0019]

[0020] in, is the actual measured value of dissolved oxygen at time k and the expected dissolved oxygen at time k difference, Indicates the difference between the current moment error and the previous moment error; is the actual measured value of nitrate nitrogen at time k and the expected nitrate nitrogen at time k difference, Indicates the difference between the current moment error and the previous moment error.

[0021] In one embodiment, determining the action space function according to the action space parameter of dissolved oxygen of the first controller and the action space of nitrate nitrogen of the first controller comprises:

[0022] Determine the action space function u according to the parameters of the first controller to be optimized a (k);

[0023] Specifically, the action space function u a (k) is the action space u of the DO agent a,DO (k) and the action space u of the NO agent a,NO (k) constitutes, satisfying u a (k)=[u a,DO (k),u a,NO (k)] T ;at the same time

[0024]

[0025] u a,NO (k) = ΔQa (k);

[0026] in, is the increment of aeration volume at time k, ΔQ a (k) is the increment of internal reflow flow.

[0027] In one embodiment, determining the reward function according to the difference between the actual measured value and the expected value of the dissolved oxygen concentration and the difference between the actual measured value and the expected value of the nitrate nitrogen concentration comprises:

[0028] The difference between the actual measured value and the expected value of the dissolved oxygen concentration is The difference between the actual measured value and the expected value of the nitrate-nitrogen concentration The reward function is r;

[0029] Among them, the reward r(k) at time k satisfies:

[0030]

[0031] Among them, δ is the threshold;

[0032] r DO =r DO = r;

[0033] Among them, the reward function of the DO agent is r DO , the reward function of the NO agent is r NO .

[0034] In one embodiment, the construction of DO agent and NO agent based on deep reinforcement learning algorithm includes:

[0035] Construct the evaluation network module of DO agent and the evaluation network module of NO agent through deep neural network;

[0036] Where: The evaluation network of the DO agent is used to fit the Q function Q(s,u a |θ1), θ1 is the network parameter, s is the input state space function, u a is the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ1);

[0037] The evaluation network of the NO agent is used to fit the Q function Q(s,u a |θ2), θ2 is the network parameter, s is the input state space function, u a is the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ2);

[0038] The evaluation network module of the DO agent includes an actual evaluation network and target evaluation network

[0039] The NO agent evaluation network module includes an actual evaluation network and target evaluation network

[0040] Constructing the action network module of the DO agent and the action network module of the NO agent through a deep neural network;

[0041] Where: The action network of the DO agent is used to fit the deterministic policy function μ(s|θ3), the network parameter is θ3, s is the input state space function, and the output is the value of the deterministic policy function μ(s|θ), while satisfying u a =μ(s|θ3),u a is the action space function;

[0042] The DO agent action network module includes an actual action network and target action network

[0043] The action network of the NO agent is used to fit the deterministic policy function μ(s|θ4), the network parameter is θ4, s is the input state space function, and the output is the value of the deterministic policy function μ(s|θ), while satisfying u a =μ(s|θ),u a is the action space function;

[0044] The NO agent action network module includes an actual action network and target action network

[0045] Construct an experience replay module for the DO agent and an experience replay module for the NO agent. The experience replay module for the DO agent is used to store the experience data generated during the training of the DO agent, and the experience replay module for the NO agent is used to store the experience data generated during the training of the NO agent.

[0046] In one embodiment, the training of the DO agent and the NO agent based on the sewage process control environment includes:

[0047] Step 4-1: Initialize the actual evaluation network of the DO agent and the real action network Parameters and And pass the parameters Initialize the target evaluation network of the DO agent Parameters Through the parameters Initialize the target action network of the DO agent Parameters

[0048] Initialize the actual evaluation network module of the NO agent and action network module Parameters and And pass the parameters Initialize the NO agent target evaluation network Parameters Through the parameters Initialize the NO agent target action network Parameters

[0049] Initialize the model learning rates of the eight networks of the DO agent and the NO agent, and update the network parameters by using the Adam gradient descent algorithm; the eight networks include the actual evaluation network of the DO agent, the target evaluation network of the DO agent, the actual action network of the DO agent, the target action network of the DO agent, the actual evaluation network of the NO agent, the target evaluation network of the NO agent, the actual action network of the NO agent, and the target action network of the NO agent;

[0050] Initialize an empty space R as the space for the experience replay module to store experience data;

[0051] Step 4-2: Set the total number of training rounds to N, the training duration of each round to D days, the current total number of training rounds n=0, and use the preset date water inflow data of the first preset number of days in dry weather provided by the simulation benchmark No. 1 model BSM1 in the activated sludge process sewage treatment process as the water inflow data of the sewage treatment system during training;

[0052] Step 4-3: Initialize the sewage treatment process control environment, specifically including steps 4-3-1 and 4-3-2:

[0053] Step 4-3-1: Initialize the biochemical reaction tank, secondary sedimentation tank and inlet water, and set the inlet water to be changed every 15 minutes;

[0054] Step 4-3-2: Initialize the first controller, specifically:

[0055] Initialize the first controller parameters and the error calculation module, where the error calculation module samples and memorizes the errors at the current moment and the previous moment, and obtains the errors corresponding to the parameters of the DO agent in the first controller by calculation. The error corresponding to the controller parameters of the NO agent in the first controller

[0056] Step 4-4: Let the initial time k = 1, the sampling time interval timestep = 15 / (60*24*20) days, and the initial expected value of dissolved oxygen concentration Expected initial nitrate nitrogen concentration

[0057] Step 4-5: Set the sampled value of dissolved oxygen concentration at time k The expected value of dissolved oxygen concentration Make a difference and get the error value of dissolved oxygen at time k The sampling value of nitrate nitrogen concentration at time k Compared with the current expected value of nitrate nitrogen concentration Make a difference and get the error value of nitrate nitrogen at time k Then and Input into the error calculation module, calculate with the error value at time k-1, and obtain a set of output values ​​x(k);

[0058] Where x(k) is composed of the output of dissolved oxygen and nitrate output Combination

[0059]

[0060] Step 4-6: Add the The actual action network fed into the DO agent The action output by the DO agent is obtained The x obtained in step 4-5 is NO,2 (k) Actual action network input into the NO agent The action output by NO agent is obtained

[0061] Step 4-7: Take the action u output by the DO agent obtained in step 4-6 a,DO (k) According to Get the aeration volume increment of the biochemical reaction tank The action u output by the NO agent obtained in step 4-6 a,NO (k) According to u a,NO (k) = ΔQa (k) Obtain the increment ΔQ of the internal reflux flow rate of the biochemical reaction tank a (k);

[0062] Step 4-8: Set the aeration volume of the fifth unit of the biochemical reaction tank to Set the internal reflux flow rate of the biochemical reaction tank to Q a (k); Then repeat Steps 4-5 at the next sampling moment to obtain the output x(k) of the error calculation module, and calculate the reward r(k) at time k according to x(k + 1) and the reward function r;

[0063] Step 4-9: Store the tuple (x(k), u a (k), r(k), x(k + 1)) into the experience replay module;

[0064] Step 4-10: Set the batch size Batch_Size = b during network model training, and the number of tuples in the current experience replay module is m. If b < m, then enter Step 4-11 to update the parameters of the eight networks in the DO agent and the NO agent; otherwise, enter Step 4-12;

[0065] Step 4-11: Randomly select b tuples from the experience replay module as training data, and then train and update the parameters of the eight networks in the DO agent and the NO agent; the update of the parameters of the eight networks satisfies Steps 4-11-1, 4-11-2, and 4-11-3:

[0066] Step 4-11-1: Take as the loss function of the actual evaluation network Q(s, u a |θ Q ), and use the gradient descent algorithm to update the parameters θ Q of the actual evaluation network:

[0067]

[0068] where, o M is the Mth state space of the agent, u aM is the Mth action of the agent, and γ = 0.99 is the discount factor;

[0069] Step 4-11-2: After updating the parameters of the actual evaluation network, then calculate the policy gradient according to and use the gradient descent algorithm to update the parameters θ μ of the actual action network μ(s|θ μ ):

[0070]

[0071] Step 4-11-3: Using the parameters of the updated actual evaluation network and actual action network and update the parameters of the target evaluation network and target action network and as follows:

[0072]

[0073] where ε is the update rate;

[0074] Step 4-12: If (k * timestep) < D, then k+1, enter Step 4-5, and continue training within this round; if (k * timestep) >= D and n < N-1, then the current total number of training rounds n+1, enter Step 4-3, and start the next round of training; if (k * timestep) >= D and n >= N-1, then training is completed, and the trained DO agent and NO agent are exported.

[0075] In one implementation, obtaining the second controller according to the trained DO agent and NO agent, and applying the second controller to the simulation model in the activated sludge process for sewage treatment, for controlling the nitrate nitrogen concentration and dissolved oxygen concentration, includes:

[0076] Subtract the sampled value of the dissolved oxygen concentration at the current sampling moment from the expected value of the current dissolved oxygen concentration to obtain the error value of the dissolved oxygen concentration at the current sampling moment

[0077] Subtract the sampled value of the nitrate nitrogen concentration at the current sampling moment from the expected value of the current dissolved oxygen concentration to obtain the error value of the dissolved oxygen concentration at the current sampling moment

[0078] Input and into the error calculation module, and calculate with the error value at the previous sampling moment to obtain a set of output quantities x(k),

[0079] where

[0080]

[0081] Input into the actual action network in the DO agent to obtain the action output by the DO agent Input x NO,2(k) Actual action network input into the NO agent The action output by NO agent is obtained

[0082] Input x(k) into the second controller corresponding to the DO agent and the NO agent, and obtain the output u of the second controller a , that is, u a,DO and u a,NO composed of u a =[u a,DO ,u a,NO ] T ;

[0083] according to The output u of the second controller a,DO Adjust the aeration volume of the fifth unit of the biochemical reaction tank According to u a,NO (k) = ΔQ a (k), the output u of the second controller a,NO Adjust the internal reflux flow Q of the biochemical reaction pool a , and then repeat the above steps at the next sampling time to control the dissolved oxygen concentration and nitrate nitrogen concentration.

[0084] A second aspect of an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method as described in any one of the above is implemented.

[0085] A third aspect of an embodiment of the present application further provides a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods listed above when executing the computer program.

[0086] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: stable tracking of the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration is achieved, and even if the expected values ​​of dissolved oxygen concentration and nitrate nitrogen concentration change, the dissolved oxygen concentration and nitrate nitrogen concentration can be quickly stabilized near the new expected values, thereby improving the accuracy of controlling dissolved oxygen concentration and nitrate nitrogen concentration. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 is an example flow chart of a multivariable control method for a municipal sewage treatment process according to an embodiment of the present application;

[0088] Figure 2 It is a schematic diagram of the WWTPs for controlling the dissolved oxygen concentration and nitrate nitrogen concentration in the wastewater treatment process of the present application;

[0089] Figure 3 This is a schematic diagram of the structure and update process of a multi-agent deep reinforcement learning DRL agent based on the DDPG algorithm;

[0090] Figure 4 It is a schematic structural diagram of the MADDPG algorithm;

[0091] Figure 5 It is a schematic diagram of the action network structure of DO agent and NO agent;

[0092] Figure 6 It is a schematic diagram of the evaluation network structure of DO agent and NO agent;

[0093] Figure 7 is a comparison diagram of the dissolved oxygen concentration control effect of the second controller of the present application;

[0094] Figure 8 This is a comparison chart of the nitrate nitrogen concentration control effect of the second controller of the present application. DETAILED DESCRIPTION

[0095] In order to make the purpose, technical solution and key points of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific implementation examples described here are only used to explain the present invention and are not used to limit this application.

[0096] Figure 1 An example flowchart provided by an embodiment of the present application is shown. For ease of explanation, only the part related to the present embodiment is shown, which is described in detail as follows.

[0097] A multivariable control method for a municipal sewage treatment process comprises steps 1 to 5 (S1-S5).

[0098] Step 1: Construct a sewage treatment process control environment based on the simulation model of the activated sludge sewage treatment process and the first controller.

[0099] Step 2: Determine the state space function according to the control errors of dissolved oxygen concentration and nitrate nitrogen concentration, determine the action space function according to the action space parameters of dissolved oxygen of the first controller and the action space of nitrate nitrogen of the first controller, and determine the reward function according to the difference between the actual measured value and the expected value of dissolved oxygen concentration and the difference between the actual measured value and the expected value of nitrate nitrogen concentration.

[0100] Step 3: Construct a DO agent and a NO agent based on a deep reinforcement learning algorithm, wherein the DO agent is used to adjust the parameters of the first controller that controls the dissolved oxygen concentration, and the NO agent is used to adjust the parameters of the first controller that controls the nitrate nitrogen concentration.

[0101] Step 4: Based on the sewage process control environment, train the DO agent and the NO agent.

[0102] Step 5: A second controller is obtained based on the trained DO agent and NO agent, and the second controller is applied to the simulation model of the activated sludge wastewater treatment process to control the nitrate nitrogen concentration and the dissolved oxygen concentration.

[0103] In this embodiment, the second controller obtained after training can stably track the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration. Even if the expected values ​​of dissolved oxygen concentration and nitrate nitrogen concentration change, the dissolved oxygen concentration and nitrate nitrogen concentration can be quickly stabilized near the new expected values, thereby improving the accuracy of controlling dissolved oxygen concentration and nitrate nitrogen concentration.

[0104] like Figure 2 As shown, in one embodiment, the simulation model and the first controller according to the activated sludge process sewage treatment process are constructed Figure 2 The sewage treatment process system WWTPs shown is used as a control environment for dissolved oxygen concentration and nitrate nitrogen concentration in the sewage treatment process, and specifically includes step 1-1 and step 1-2.

[0105] Step 1-1: Construct a biochemical reaction tank and a secondary sedimentation tank according to the simulation benchmark model No. 1 BSM1 in the activated sludge method sewage treatment process; the biochemical reaction tank includes the first unit, the second unit, the third unit, the fourth unit and the fifth unit; the first unit and the second unit are anoxic zones, which mainly carry out the denitrification reaction process; in anoxic environments, denitrifying bacteria use organic carbon and nitrate as donors to reduce nitrate nitrogen in the treatment process to nitrogen gas. The third unit, the fourth unit and the fifth unit are aerobic zones, which mainly carry out the nitrification reaction process; in an aerobic environment, nitrifying bacteria oxidize ammonia nitrogen in the treatment process into nitrate. The secondary sedimentation tank is divided into 10 layers, which mainly carry out the mud-water separation process. The supernatant after sewage treatment is discharged into the receiving water body through the upper layer of the secondary sedimentation tank, while the sludge is discharged through the lower layer. A part of it is treated as residual sludge, and the other part is returned to the biochemical reaction tank as a reaction carrier through external reflux to continue to participate in the biological treatment process of sewage.

[0106] Step 1-2: construct a first controller as a dissolved oxygen concentration controller of the fifth unit of the biochemical reaction tank and a nitrate nitrogen concentration controller of the second unit of the biochemical reaction tank; wherein the first controller is a MADRL controller.

[0107] Among them, the first controller is a MADRL controller.

[0108] In one embodiment, step 2-1: determining the state space function according to the control error of dissolved oxygen concentration and nitrate nitrogen concentration comprises:

[0109] The state space function s(k) is the state space o of the DO agent. DO (k) and the state space o of the NO agent NO (k) is composed of, satisfying s(k) = [o DO (k) T ,o NO (k) T ] T ;

[0110] The state space o of the DO agent DO (k) Control error by dissolved oxygen concentration constitute, satisfy

[0111] The state space of the NO agent is controlled by the nitrate nitrogen concentration error constitute, satisfy at the same time

[0112]

[0113] in, is the actual measured value of dissolved oxygen at time k and the expected dissolved oxygen at time k difference, Indicates the difference between the current moment error and the previous moment error; is the actual measured value of nitrate nitrogen at time k and the expected nitrate nitrogen at time k difference, Indicates the difference between the current moment error and the previous moment error.

[0114] In one embodiment, step 2-2: determining the action space function according to the action space parameter of dissolved oxygen of the first controller and the action space of nitrate nitrogen of the first controller comprises:

[0115] Determine the action space function u according to the parameters of the first controller to be optimized a (k);

[0116] Specifically, the action space function u a (k) is the action space u of the DO agent a,DO (k) and the action space u of the NO agent a,NO (k) constitutes, satisfying u a (k)=[u a,DO (k),u a,NO (k)] T ;at the same time

[0117]

[0118] u a,NO (k) = ΔQ a (k);

[0119] in, is the increment of aeration volume at time k, ΔQ a (k) is the increment of internal reflow flow.

[0120] In one embodiment, step 2-3: determining the reward function according to the difference between the actual measured value and the expected value of the dissolved oxygen concentration and the difference between the actual measured value and the expected value of the nitrate nitrogen concentration, comprises:

[0121] The difference between the actual measured value and the expected value of the dissolved oxygen concentration is The difference between the actual measured value and the expected value of the nitrate-nitrogen concentration The reward function is r;

[0122] Among them, the reward r(k) at time k satisfies:

[0123]

[0124] Among them, δ is the threshold;

[0125] r DO =r NO = r;

[0126] Among them, the reward function of the DO agent is r DO , the reward function of the NO agent is r NO , both are calculated in the same way as r.

[0127] Combination Figure 3 and Figure 4 As shown, in one embodiment, according to Figure 3 The agent model based on the DDPG algorithm shown and Figure 4 The MADDPG algorithm block diagram shown builds the DO agent and NO agent models based on the MADDPG algorithm. Specifically, step 3 includes step 3-1, step 3-2 and step 3-3.

[0128] Step 3-1: Construct the evaluation network module of DO agent and the evaluation network module of NO agent through deep neural network;

[0129] Where: The evaluation network of the DO agent is used to fit the Q function Q(s,u a |θ1), θ1 is the parameter of the network, s is the state space function of the input, u ais the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ1);

[0130] The evaluation network of the NO agent is used to fit the Q function Q(s,u a |θ2), θ2 is the network parameter, s is the input state space function, u a is the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ2);

[0131] The evaluation network module of the DO agent includes an actual evaluation network and target evaluation network

[0132] The NO agent evaluation network module includes an actual evaluation network and target evaluation network

[0133] like Figure 5 As shown in the figure, the actual evaluation network and the target evaluation network have the same structure, both of which contain four fully connected network layers, namely one input layer, two hidden layers and one output layer; the hidden layers all use the ReLU function as the activation function, and the number of hidden layer nodes is 128.

[0134] Step 3-2: constructing the action network module of the DO agent and the action network module of the NO agent through a deep neural network;

[0135] Where: The action network of the DO agent is used to fit the deterministic policy function μ(s|θ3), the network parameter is θ3, s is the input state space function, and the output is the value of the deterministic policy function μ(s|θ), while satisfying u a =μ(s|θ3),u a is the action space function;

[0136] The DO agent action network module includes an actual action network and target action network

[0137] The action network of the NO agent is used to fit the deterministic policy function μ(s|θ4), the network parameter is θ4, s is the input state space function, and the output is the value of the deterministic policy function μ(s|θ), while satisfying u a =μ(s|θ),u a is the action space function;

[0138] The NO agent action network module includes an actual action network and target action network like Figure 6 As shown in the figure, the actual action network and target action network of DO agent and NO agent have the same structure, both of which contain four fully connected network layers, namely one input layer, two hidden layers and one output layer; the hidden layers all use ReLU function as the activation function, and the number of nodes in the hidden layer is 128, while the output layers all use Tanh function as the activation function.

[0139] Step 3-3: Construct the experience replay module of the DO agent and the experience replay module of the NO agent, the experience replay module of the DO agent is used to store the experience data generated during the training of the DO agent, and the experience replay module of the NO agent is used to store the experience data generated during the training of the NO agent.

[0140] In one embodiment, step 4 includes steps 4-1 to 4-12.

[0141] Step 4-1: Initialize the actual evaluation network of the DO agent and the real action network Parameters and And pass the parameters Initialize the target evaluation network of the DO agent Parameters By the parameters Initialize the target action network of the DO agent Parameters

[0142] Initialize the actual evaluation network module of the NO agent and action network module Parameters and And pass the parameters Initialize the NO agent target evaluation network Parameters By the parameters Initialize the NO agent target action network Parameters

[0143] Initialize the model learning rates of the eight networks of the DO agent and the NO agent, and update the network parameters by using the Adam gradient descent algorithm; the eight networks include the actual evaluation network of the DO agent, the target evaluation network of the DO agent, the actual action network of the DO agent, the target action network of the DO agent, the actual evaluation network of the NO agent, the target evaluation network of the NO agent, the actual action network of the NO agent, and the target action network of the NO agent. In one embodiment, the learning rate of the action network is 1*10 -4 , the learning rate of the evaluation network is 3*10 -4 .

[0144] Initialize an empty space R as a space for the experience playback module to store experience data. In one embodiment, the capacity of the empty space R is 4*10 5 .

[0145] Step 4-2: Set the total number of training rounds to N, the training duration of each round to D days, the current total number of training rounds n = 0, and use the preset date water inflow data of the first preset days of dry weather provided by the simulation benchmark model No. 1 BSM1 in the activated sludge process as the water inflow data of the sewage treatment system (WWTPs) during training. In one embodiment, N is 200, D is 1, the first preset days are 14 days, and the preset date is the first 7 days of the first preset days.

[0146] Step 4-3: Initialize the sewage treatment process control environment, specifically including steps 4-3-1 and 4-3-2:

[0147] Step 4-3-1: Initialize the biochemical reaction tank, secondary sedimentation tank and inlet water, and set the inlet water to be changed every 15 minutes;

[0148] Step 4-3-2: Initialize the first controller, specifically:

[0149] Initialize the first controller parameters and the error calculation module, where the error calculation module samples and memorizes the errors at the current moment and the previous moment, and obtains the errors corresponding to the parameters of the DO agent in the first controller by calculation. The error corresponding to the controller parameters of the NO agent in the first controller Among them, the error between the current moment and the previous moment is 0.

[0150] Step 4-4: Let the initial time k = 1, the sampling time interval timestep = 15 / (60*24*20) days, and the initial expected value of dissolved oxygen concentration Expected initial nitrate nitrogen concentration

[0151] Step 4-5: Set the sampled value of dissolved oxygen concentration at time k The expected value of dissolved oxygen concentration Make a difference and get the error value of dissolved oxygen at time k The sampling value of nitrate nitrogen concentration at time k Compared with the current expected value of nitrate nitrogen concentration Make a difference and get the error value of nitrate nitrogen at time k Then and Input into the error calculation module, calculate with the error value at time k-1, and obtain a set of output values ​​x(k);

[0152] Where x(k) is composed of the output of dissolved oxygen and nitrate output Combination

[0153]

[0154] Step 4-6: Add the The actual action network fed into the DO agent The action output by the DO agent is obtained The x obtained in step 4-5 is NO,2 (k) Actual action network input into the NO agent The action output by NO agent is obtained

[0155] Step 4-7: Take the action u output by the DO agent obtained in step 4-6 a,DO (k) According to Get the aeration volume increment of the biochemical reaction tank The action u output by the NO agent obtained in step 4-6 a,NO (k) According to u a,NO (k) = ΔQ a (k) Obtain the internal reflux flow increment ΔQ of the biochemical reaction pool a (k).

[0156] Step 4-8: Set the aeration rate of the fifth unit of the biochemical reaction tank to The internal reflux flow rate of the biochemical reaction pool is set to Q a (k); then repeat steps 4-5 at the next sampling moment to obtain the output x(k) of the error calculation module, and calculate the reward r(k) at moment k based on x(k+1) and the reward function r.

[0157] Step 4-9: Substitute the tuple (x(k),ua (k), r(k), x(k + 1)) is stored into the experience replay module.

[0158] Step 4 - 10: Set the batch size Batch_Size = b during network model training. The number of tuples in the current experience replay module is m. If b < m, then go to Step 4 - 11 to update the parameters of the eight networks in the DO agent and the NO agent; otherwise, go to Step 4 - 12.

[0159] Step 4 - 11: Randomly select b tuples from the experience replay module as training data, and then train and update the parameters of the eight networks in the DO agent and the NO agent; the update of the parameters of the eight networks satisfies Step 4 - 11 - 1, Step 4 - 11 - 2, and Step 4 - 11 - 3.

[0160] Step 4 - 11 - 1: Take as the loss function of the actual evaluation network Q(s, u a |θ Q ), and use the gradient descent algorithm to update the parameters θ Q of the actual evaluation network:

[0161]

[0162] where, o M is the M - th state space of the agent, u aM is the M - th action of the agent, and γ = 0.99 is the discount factor;

[0163] Step 4 - 11 - 2: After updating the parameters of the actual evaluation network, calculate the policy gradient according to , and use the gradient descent algorithm to update the parameters θ μ of the actual action network μ(s|θ μ ):

[0164]

[0165] Step 4 - 11 - 3: Use the updated parameters and of the actual evaluation network and the actual action network to update the parameters and of the target evaluation network and the target action network:

[0166]

[0167] where, ε is the update rate;

[0168] Step 4-12: If (k * timestep) < D, then k+1, go to Step 4-5, and continue training in this round; if (k * timestep) >= D and n < N-1, then the current total number of training rounds n+1, go to Step 4-3, and start the next round of training; if (k * timestep) >= D and n >= N-1, then the training is completed, and the trained DO agent and NO agent are exported.

[0169] In one implementation, the trained second controller is obtained by combining the trained DO agent and NO agent, and the second controller is applied to the BSM1 benchmark simulation platform in the activated sludge process for sewage treatment to realize the coordinated control of the dissolved oxygen concentration in the fifth unit and the nitrate nitrogen concentration in the second unit in the biochemical reaction tank. The specific control process includes Step 5-1 to Step 5-4.

[0170] Step 5-1: Take the sampled value of the dissolved oxygen concentration at the current sampling moment and subtract it from the current expected value of the dissolved oxygen concentration to obtain the error value of the dissolved oxygen concentration at the current sampling moment

[0171] Take the sampled value of the nitrate nitrogen concentration at the current sampling moment and subtract it from the current expected value of the dissolved oxygen concentration to obtain the error value of the dissolved oxygen concentration at the current sampling moment

[0172] Input and into the error calculation module, and calculate them with the error value at the previous sampling moment to obtain a set of output quantities x(k).

[0173] Among them,

[0174]

[0175] Step 5-2: Input into the actual action network in the DO agent to obtain the action output by the DO agent. Input x NO,2 (k) into the actual action network in the NO agent to obtain the action

[0176] Step 5-3: Input x(k) into the second controllers corresponding to the DO agent and the NO agent to obtain the output u a of the second controller, that is, from ua,DO and u a,NO composed of u a =[u a,DO ,u a,NO ] T .

[0177] Step 5-4: According to The output u of the second controller a,DO Adjust the aeration volume of the fifth unit of the biochemical reaction tank According to u a,NO (k) = ΔQ a (k), the output u of the second controller a,NO Adjust the internal reflux flow Q of the biochemical reaction pool a , and then repeat the above steps at the next sampling time to control the dissolved oxygen concentration and nitrate nitrogen concentration.

[0178] See also Figure 7 and Figure 8 , it can be seen that the second controller of the present application can stably track the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration, and even if the expected values ​​of dissolved oxygen concentration and nitrate nitrogen concentration change, the dissolved oxygen concentration and nitrate nitrogen concentration can be quickly stabilized near the new expected values, and compared with the PID controller, better tracking effect and smaller tracking error can be achieved. Based on this result, it can be seen that the multivariable control method for urban sewage treatment process provided by the present application has better adaptive ability than the traditional PID controller, and can also quickly and accurately control the dissolved oxygen concentration and nitrate nitrogen concentration.

[0179] This application adopts deep reinforcement learning technology, and applies it to the multivariable control system of the urban sewage treatment process in view of the nonlinearity, time-varying and significant hysteresis characteristics of the sewage treatment process. The deep reinforcement learning technology can adapt to the changes of the uncertain system through interaction with the environment and self-learning, thereby improving the adaptability and adaptive control ability of the MADRL controller in the face of the complex and changing environment in the sewage treatment process. This application also uses a multi-agent deep reinforcement learning algorithm to coordinate the concentrations of dissolved oxygen DO and nitrate nitrogen NO in the urban sewage treatment process, effectively coping with the control difficulties of nonlinear systems. Not only does it achieve accurate real-time control of dissolved oxygen DO and nitrate nitrogen NO concentrations, it also overcomes the problems of insufficient accuracy and poor stability of traditional PID control strategies in multivariable control of sewage treatment processes.

[0180] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method as described in any one of the above is implemented.

[0181] An embodiment of the present application further provides a server, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods listed above when executing the computer program.

[0182] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0183] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0184] Those of ordinary skill in the art will appreciate that the elements and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0185] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A multivariable control method for urban sewage treatment process, It is characterized in that include: Constructing a sewage treatment process control environment according to a simulation model of an activated sludge method sewage treatment process and a first controller; determining a state space function according to control errors of dissolved oxygen concentration and nitrate nitrogen concentration, determining an action space function according to action space parameters of dissolved oxygen of the first controller and action space parameters of nitrate nitrogen of the first controller, and determining a reward function according to a difference between an actual measured value and an expected value of dissolved oxygen concentration and a difference between an actual measured value and an expected value of nitrate nitrogen concentration; Constructing a DO agent and a NO agent based on a deep reinforcement learning algorithm, wherein the DO agent is used to adjust the parameters of the first controller for controlling the dissolved oxygen concentration, and the NO agent is used to adjust the parameters of the first controller for controlling the nitrate nitrogen concentration; Based on the sewage process control environment, training the DO agent and the NO agent; as well as A second controller is obtained according to the trained DO agent and NO agent, and the second controller is applied to the simulation model of the activated sludge wastewater treatment process to control the nitrate nitrogen concentration and the dissolved oxygen concentration.

2. The method according to claim 1, It is characterized in that The construction of a sewage treatment process control environment based on the simulation model of the activated sludge sewage treatment process and the first controller includes: According to the simulation benchmark model No. 1 BSM1 in the activated sludge wastewater treatment process, a biochemical reaction tank and a secondary sedimentation tank are constructed; wherein the biochemical reaction tank includes a first unit, a second unit, a third unit, a fourth unit and a fifth unit; A first controller is constructed as a dissolved oxygen concentration controller of the fifth unit of the biochemical reaction tank and a nitrate nitrogen concentration controller of the second unit of the biochemical reaction tank; wherein the first controller is a MADRL controller.

3. The method according to claim 1, It is characterized in that The step of determining the state space function according to the control errors of the dissolved oxygen concentration and the nitrate nitrogen concentration comprises: The state space function s(k) is the state space o of the DO agent. DO (k) and the state space o of the NO agent DO (k) is composed of, satisfying s(k) = [o DO (k) T ,o DO (k) T ] T ; The state space o of the DO agent DO (k) Error controlled by dissolved oxygen concentration (k) constitute, satisfy The state space of the NO agent is controlled by the nitrate nitrogen concentration error constitute, satisfy at the same time in, is the actual measured value of dissolved oxygen at time k and the expected dissolved oxygen at time k difference, Indicates the difference between the current moment error and the previous moment error; is the actual measured value of nitrate nitrogen at time k and the expected nitrate nitrogen at time k difference, Indicates the difference between the current moment error and the previous moment error.

4. The method according to claim 1, It is characterized in that The determining of the action space function according to the action space parameter of dissolved oxygen of the first controller and the action space of nitrate nitrogen of the first controller comprises: Determine the action space function u according to the parameters of the first controller to be optimized a (k); Specifically, the action space function u a (k) is the action space u of the DO agent a,DO (k) and the action space u of the NO agent a,NO (k) constitutes, satisfying u a (k)=[u a,DO (k),u a,NO (k)] T ;at the same time in, is the increment of aeration volume at time k, ΔQ a (k) is the increment of internal reflow flow.

5. The method according to claim 1, It is characterized in that The step of determining a reward function according to a difference between an actual measured value and an expected value of dissolved oxygen concentration and a difference between an actual measured value and an expected value of nitrate nitrogen concentration comprises: The difference between the actual measured value and the expected value of the dissolved oxygen concentration is The difference between the actual measured value and the expected value of the nitrate-nitrogen concentration The reward function is r; Among them, the reward r(k) at time k satisfies: Among them, δ is the threshold; r DO =r DO =r; Among them, the reward function of the DO agent is r DO , the reward function of the NO agent is r DO .

6. The method according to claim 1, It is characterized in that The method of constructing a DO agent and a NO agent based on a deep reinforcement learning algorithm includes: Construct the evaluation network module of DO agent and the evaluation network module of NO agent through deep neural network; Where: The evaluation network of the DO agent is used to fit the Q function Q(s,u a |θ1), θ1 is the parameter of the network, s is the state space function of the input, u a is the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ1); The evaluation network of the NO agent is used to fit the Q function Q(s,u a |θ2), θ2 is the network parameter, s is the input state space function, u a is the input action space function, Q is the value of the output function, satisfying Q=Q(s,u a |θ2); The evaluation network module of the DO agent includes an actual evaluation network and target evaluation network The NO agent evaluation network module includes an actual evaluation network and target evaluation network Constructing the action network module of the DO agent and the action network module of the NO agent through a deep neural network; Where: the action network of the DO agent is used to fit the deterministic policy function μ(s|θ3), the network parameter is θ3, s is the input state space function, the output is the value of the deterministic policy function μ(s|θ), and u a =μ(s|θ3),u a is the action space function; The DO agent action network module includes an actual action network and target action network The action network of the NO agent is used to fit the deterministic policy function μ(s|θ4), the network parameter is θ4, s is the input state space function, and the output is the value of the deterministic policy function μ(s|θ), while satisfying u a =μ(s|θ),u a is the action space function; The NO agent action network module includes an actual action network and target action network Construct the experience replay module of the DO agent and the experience replay module of the NO agent. The experience replay module of the DO agent is used to store the experience data generated during the training of the DO agent, and the experience replay module of the NO agent is used to store the experience data generated during the training of the NO agent.

7. The method according to claim 6, wherein, training the DO agent and the NO agent based on the sewage process control environment includes: Step 4-1: Initialize the actual evaluation network of the DO agent and the real action network Parameters and And pass the parameters Initialize the target evaluation network of the DO agent Parameters Through the parameters Initialize the target action network of the DO agent Parameters Initialize the actual evaluation network module of the NO agent and action network module Parameters and And pass the parameters Initialize the NO agent target evaluation network Parameters Through the parameters Initialize the NO agent target action network Parameters Initializing the model learning rates of the eight networks of the DO agent and the NO agent, and updating the network parameters through the Adam gradient descent algorithm; the eight networks include the actual evaluation network of the DO agent, the target evaluation network of the DO agent, the actual action network of the DO agent, the target action network of the DO agent, the actual evaluation network of the NO agent, the target evaluation network of the NO agent, the actual action network of the NO agent, and the target action network of the NO agent; Initialize an empty space R as the space for the experience replay module to store experience data; Step 4-2: Set the total number of training rounds to N, the training duration per round to D days, the current total number of training rounds n = 0, and use the preset date inlet water data in the first preset number of days of inlet water data under dry weather provided by the simulation benchmark model BSM1 in the activated sludge process sewage treatment as the inlet water data of the sewage treatment system during training; Step 4-3: Initialize the sewage process control environment, specifically including steps 4-3-1 and 4-3-2: Step 4-3-1: Initialize the biochemical reaction tank, the secondary sedimentation tank, and the inlet water, and set the inlet water to change every 15 minutes; Step 4-3-2: Initialize the first controller, specifically: Initialize the first controller parameters and the error calculation module, where the error calculation module samples and memorizes the errors at the current moment and the previous moment, and obtains the errors corresponding to the parameters of the DO agent in the first controller by calculation. The error corresponding to the controller parameters of the NO agent in the first controller Step 4-4: Let the initial time k = 1, the sampling time interval timestep = 15 / (60*24*20) days, and the initial expected value of dissolved oxygen concentration Expected initial nitrate nitrogen concentration Step 4-5: Set the sampled value of dissolved oxygen concentration at time k The expected value of dissolved oxygen concentration Make a difference and get the error value of dissolved oxygen at time k The sampling value of nitrate nitrogen concentration at time k Compared with the current expected value of nitrate nitrogen concentration Make a difference and get the error value of nitrate nitrogen at time k Then and Input into the error calculation module, calculate with the error value at time k-1, and obtain a set of output values ​​x(k); Where x(k) is composed of the output of dissolved oxygen and nitrate output Combination Step 4-6: Add the The actual action network fed into the DO agent The action output by the DO agent is obtained The x obtained in step 4-5 is NO,2 (k) Actual action network input into the NO agent The action output by NO agent is obtained Step 4-7: Take the action u output by the DO agent obtained in step 4-6 a,DO (k) According to Get the aeration volume increment of the biochemical reaction tank The action u output by the NO agent obtained in step 4-6 a,NO (k) According to u a,NO (k) = ΔQ a (k) Obtain the internal reflux flow increment ΔQ of the biochemical reaction pool a (k); Step 4-8: Set the aeration rate of the fifth unit of the biochemical reaction tank to The internal reflux flow rate of the biochemical reaction pool is set to Q a (k); then repeat steps 4-5 at the next sampling time to obtain the output x(k) of the error calculation module, and calculate the reward r(k) at time k based on x(k+1) and the reward function r; Step 4-9: Substitute the tuple (x(k),u a (k), r(k), x(k+1)) are stored in the experience playback module; Step 4-10: Set the batch size Batch_Size = b during network model training, and the number of tuples in the current experience replay module is m. If b < m, then enter Step 4-11 to update the parameters of the eight networks in the DO agent and the NO agent, otherwise enter Step 4-12; Step 4-11: Randomly select b tuples from the experience replay module as training data, and then train and update the parameters of the eight networks in the DO agent and the NO agent; the update of the parameters of the eight networks satisfies Steps 4-11-1, 4-11-2, and 4-11-3: Step 4-11-1: As the actual evaluation network Q(s,u a |θ Q ) loss function, and use the gradient descent algorithm to update the parameters θ of the actual evaluation network Q : in, o M is the Mth state space of the agent, u aM is the Mth action of the agent, γ = 0.99 is the discount factor; Step 4-11-2: After updating the parameters of the actual evaluation network, Calculate the policy gradient and use the gradient descent algorithm to update the actual action network μ(s|θ μ ) parameter θ μ : Step 4-11-3: Use the updated parameters of the actual evaluation network and the actual action network and Parameters of the target evaluation network and the target action network and To update: where ε is the update rate; Step 4-12: If (k * timestep) < D, then k + 1, enter Step 4-5, and continue training within this round; if (k * timestep) >= D and n < N - 1, then the current total number of training rounds n + 1, enter Step 4-3, and start the next round of training; if (k * timestep) >= D and n >= N - 1, then the training is completed, and the trained DO agent and NO agent are exported.

8. The method according to claim 7, wherein, The second controller is obtained according to the trained DO agent and NO agent, and the second controller is applied to the simulation model of the activated sludge wastewater treatment process to control the nitrate nitrogen concentration and the dissolved oxygen concentration, including: The sampled value of dissolved oxygen concentration at the current sampling time The expected value of dissolved oxygen concentration Make a difference and get the error value of dissolved oxygen concentration at the current sampling time The sampling value of nitrate nitrogen concentration at the current sampling time The expected value of dissolved oxygen concentration Make a difference and get the error value of dissolved oxygen concentration at the current sampling time Will and Input into the error calculation module, calculate with the error value of the previous sampling moment, and get a set of output quantities x(k). in, Will The actual action network fed into the DO agent The action output by the DO agent is obtained x NO,2 (k) Actual action network input into the NO agent The action output by NO agent is obtained Input x(k) into the second controller corresponding to the DO agent and the NO agent, and obtain the output u of the second controller a , that is, u a,DO and u a,NO composed of u a =[u a,DO ,u a,NO ] T ; according to The output u of the second controller a,DO Adjust the aeration volume of the fifth unit of the biochemical reaction tank According to u a,NO (k) = ΔQ a (k), the output u of the second controller a,NO Adjust the internal reflux flow Q of the biochemical reaction pool a , and then repeat the above steps at the next sampling time to control the dissolved oxygen concentration and nitrate nitrogen concentration.

9. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

10. A server comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Citation Information

Cited By

  • Intelligent control method and system for sewage treatment

    CN119620623A

  • Intelligent factory multi-device cooperative control method based on mutual information driven sparse reward

    CN120652939A